OpenGradient

OpenGradient is a decentralized Layer 1 blockchain purpose-built for verifiable AI inference. Every computation on the network is cryptographically verifiable without trusting any single party: models run on a permissionless network of specialized nodes, proofs are settled on-chain, and the entire pipeline from request to response is auditable. The project raised $8.5 million from a16z Crypto, Coinbase Ventures, SV Angel, and Foresight Ventures, with angel investors including Balaji Srinivasan, Illia Polosukhin, and Sandeep Nailwal.

Key Features

• Verifiable inference by default: Developers choose between ZKML (zero-knowledge proofs, mathematically certain but 1000-10000x slower), TEE (hardware-attested enclaves with negligible overhead), and Vanilla (signature-only) verification.
• HACA architecture: The Hybrid AI Compute Architecture separates execution from verification. Inference nodes run models with web2-like latency; full nodes verify proofs and settle them in consensus asynchronously.
• PIPE engine: The Parallelized Inference Pre-Execution Engine runs ML models with atomic guarantees, so inference results can be part of an onchain state transition rather than an external oracle call.
• x402 LLM execution: Payment-gated access to OpenAI, Anthropic, Google, xAI, and more through TEE-verified infrastructure, settled on-chain and paid with OPG tokens.
• Model Hub: A censorship-resistant, decentralized model repository on Walrus, positioning itself as a community-owned alternative to Hugging Face, with 2,000+ models hosted on mainnet.
• Python SDK: The primary interface (pip install opengradient) for LLM inference, model management, and Model Hub integration.

Performance & Scale

As of mid-2026, the network reports over 2 million verifiable inferences processed, more than 500,000 cryptographic proofs verified, and over 263,500 unique wallets interacting with the system. The Nova Testnet launched October 1, 2025, embedding AI computation and its proof directly into consensus; the OPG token generation event followed on April 21, 2026.

Use Cases

• DeFi with AI risk models: Use ZKML-verified ML models for liquidations, credit scoring, and other high-stakes financial decisions that must be provably correct.
• Verifiable LLM services: Access GPT-4, Claude, Grok, and Gemini through a unified API where every inference is TEE-verified, useful for audit trails, compliance, and dispute resolution.
• Agents with provable reasoning: Build autonomous agents where every LLM call is cryptographically signed with the exact prompt used, so anyone can verify the reasoning chain onchain.
• Atomic AI transactions: Run inference as part of a smart contract state transition, so the result cannot be separated from the transaction.
• Composable AI workflows: Chain multiple models with mixed verification methods in a single transaction, e.g., ZKML for the risk model and TEE for the LLM reasoning.

Getting Started

1. Install the SDK: pip install opengradient.
2. Create an account and obtain API credentials from the OpenGradient dashboard.
3. Upload a model to the Model Hub or call a hosted model via the inference API.
4. Choose a verification method (ZKML, TEE, or Vanilla) based on your risk profile.
5. For onchain use, deploy with Ethers.js or the Verifiable Inference SDK against the OpenGradient network (testnet RPC: https://eth-devnet.opengradient.ai, chain ID 10744).

Advantages & Comparison

Traditional centralized AI APIs are black boxes: you must trust the provider's logs about what model ran and what it produced. OpenGradient replaces that trust with cryptographic evidence. Unlike general-purpose blockchains that bolt AI on as an oracle, OpenGradient makes inference a first-class citizen of blockspace, with execution and proof sharing a single Merkle root. The main trade-off is cost: ZKML verification is expensive for large models, and latency is higher than calling a centralized API directly, though the network's web2-like response path mitigates this.

Frequently Asked Questions

How much does it cost?

Inference is metered and paid with OPG tokens. ZKML verification carries heavy computational overhead (1000-10000x vanilla); TEE and Vanilla verification are much cheaper. Exact rates are listed on the network's pricing page.

Is OpenGradient open source?

The core SDK, examples, and several network components are open source on the OpenGradient GitHub organization.

Do I need crypto to use it?

For LLM execution via x402, payments settle onchain with OPG tokens. The Python SDK also supports non-custodial onboarding flows for builders who want API-style access.

Alternatives

• OpenRouter: A simpler unified API for many models without onchain verification: /en/items/ai-ecosystem/openrouter
• Langfuse: LLM observability and tracing for verification via logging: /en/items/agent-infra/langfuse
• ToAPIs: An OpenAI-compatible API gateway with routing and failover: /en/items/ai-ecosystem/toapis

Conclusion

OpenGradient is the most ambitious attempt yet to make AI inference verifiable by default, combining a dedicated Layer 1, specialized compute nodes, and a choice of cryptographic and hardware verification. Teams building high-stakes onchain agents, DeFi risk models, or audit-grade AI services should evaluate it; everyone else can start with its TEE-verified LLM API and grow into stronger guarantees as needs demand.

Usage Terms

Use of the network is subject to OpenGradient's Terms of Use and the OPG token economics described in the project's documentation.